"Just as we now glance at nutrition labels without a second thought, we could one day check content labels before diving into a news article or blog post." Column: Why the internet needs content labels 📝 Tali Sharot and Christopher A. Kelly time.com/7177598/intern…

AI is moving so fast and it's dominated by industry. The CREATE AI Act will drive public-interest AI research, making sure that AI benefits everyone, says Stanford HAI executive director Russell Wald. Will Congress push it through? Read this Q&A: stanford.io/417WutE

I don't really see a clear path where we keep an open internet that is not mostly full of AIs talking to each other. We can't reliably detect AI content, it is cheap and easy to generate, and there are lots of incentives to do so. You can see the problem on all the social sites.

If AI development stopped this week we would have 5-10 years of absorbing the impact of current models on education, culture, healthcare, and business. But this week has also suggested that development is not stopping.

Academic Christmas

.Moshe Glickman and affective.brain reveal a human-AI feedback loop, where AI amplifies subtle human biases, which are then further internalized by humans. This cycle increases human bias over time across domains. nature.com/articles/s4156…

Are we more predators or prey? We’ve guessed wrong for a century. Today we can shoot wolves from helicopters but—I argue in the The New York Times today—humans evolved more as hunted than hunters. Link ⬇️

"The ChatGPT moment for general robotics is just around the corner," said Nvidia CEO Jensen Huang during his #CES2025 keynote presentation.

The US Copyright office has ruled that AI/human combined work can be copyrighted as long as a human is adding, changing or selecting elements. Prompts alone do not usually produce copyrighted work. Everything is case-by-case, but the report is clear and thoughtful about reasoning

Fun to watch Deep Research basically go through the same learning journey I did in my first year PhD classes on organizational theory in order to answer a question, though this was in a few seconds, rather than a semester (and I hope I grappled with this more deeply myself?)

The amount of capability overhang in current AI systems is hard to overstate, even in narrow areas like vision & image creation. If AI development stopped today (and no indication that is happening), we have a couple decades of figuring out how to integrate it into work & life.

🧵 1/ How well do LLMs actually do on Olympiad-level math? We evaluated frontier models on 455 problems from the IMO Shortlist. Unlike most benchmarks, we emphasize proof validity, not just final answer correctness. Here’s what we found 👇

Anthropic continues to put out solid studies of how Claude is used, this time in universities It shows both AI perils & promise. Students are using it for all sorts of purposes, with getting direct answers making up a lot use, but many cases of AI supporting their learning; too.

With Google's AlphaEvolve, we have evidence that LLMs can discover novel & useful ideas, when put together with the right tooling. These results are impressive: given 50 open math problems, the AI rediscovered the leading approach 75% of the time & improved on it 20% of the time

New paper in Nature Human Behavior (led by Matan Rubin), where we manipulated the perceived identity of a real-time empathic response to be or AI. We observed a drop in perceived empathy when label was AI, and this effect was mainly driven by emotion. rdcu.be/et9ad

We'll be talking about advice, feedback, and conversations. If you're at #IACM2025, then check out our symposium this coming Monday! IACM (🧵)

"How much time do I have left? I don't mean for this presentation, I mean as a professional economist." -Joel Waldfogel discussing a paper at our NBER Workshop on the Economics of AI last week 😱

Thanks to Institute for New Economic Thinking for having me! Excited to see more bridges being built between the CS and Economics communities.

📣 A new Stanford Digital Economy Lab study shows one of the first large-scale evidence of genAI's impact on entry-level workers. Using data from ADP, the study reveals a sharp decline in employment in AI-exposed occupations, but not in jobs where AI augments human work. digitaleconomy.stanford.edu/publications/c…

I had access to Gemini 3. It is a very good, very fast model. It also demonstrates the change from chatbot to agent. oneusefulthing.org/p/three-years-…